The crystal structure and biochemical characterization of Kif15: a bifunctional molecular motor involved in bipolar spindle formation and neuronal development

The structural and biochemical study of Kif15 provides insight into this potential drug target and allows comparison with Eg5, a kinesin that partially shares the functions of Kif15.

Kinesins constitute a superfamily of microtubule-based motor proteins with important cellular functions ranging from intracellular transport to cell division. Some kinesin family members function during the mitotic phase of the eukaryotic cell cycle and are crucial for the successful progression of cell division. In the early stages of mitosis, during prometaphase, certain kinesins are required for the formation of the bipolar spindle, such as Eg5 and Kif15, which seem to possess partially overlapping functions. Because kinesins transform the chemical energy from ATP hydrolysis into mechanical work, inhibition of their function is a tractable approach for drug development. Drugs targeting Eg5 have shown promise as anticancer agents. Kif15 has recently come to the fore because it can substitute the functions of Eg5, and may itself have potential as a prospective drug target. Here, the initial biochemical, kinetic and structural characterization of Kif15 is reported and it is compared with the functionally related motor Eg5. Although Kif15 contains ADP in the catalytic site, its motor-domain structure was captured in the 'ATP-like' configuration, with the neck linker docked to the catalytic core. The interaction of Kif15 with microtubules was also investigated and structural differences between these two motors were elucidated which indicate profound differences in their mode of action, in agreement with current models of microtubule cross-linking and sliding.
The involvement of Eg5 in the separation of the duplicated centrosomes (Blangy et al., 1995) has made it a potential target for drug development in cancer chemotherapy, with several kinesin-5-specific inhibitors in multiple Phase I and II clinical trials (Huszar et al., 2009;Rath & Kozielski, 2012). Interestingly, Kif15 overexpression has been shown to be able to restore functions of Eg5 under certain conditions, for example when Eg5 is depleted (Tanenbaum et al., 2009). Subsequently, it has been hypothesized (although not yet proven) that tumours treated with Eg5-targeting drugs might acquire resistance to these inhibitors by simply up-regulating Kif15. This interesting hypothesis and the involvement of Kif15 in bipolar spindle assembly makes it a potential target for drug development in cancer chemotherapy, and the first inhibitor scaffolds targeting Kif15 have been reported in the patent literature (McDonald et al., 2004).
Although the role of Kif15 in bipolar spindle formation has been investigated in a variety of organisms (Boleti et al., 1996;Rogers et al., 2000;Segbert et al., 2003;Sueishi et al., 2000) and has been compared with that of Eg5 in cellular assays, much less is known about its biochemical, mechanochemical and structural properties or its interactions with binding partners. The recombinant expression of human Kif15 for inhibitor screening, and the structure determination of its motor domain for structure-based design, would be highly desirable to facilitate the drug-development process. Here, we provide the initial biochemical characterization of human Kif15 and compare its properties with those of its functionally related partner Eg5. We reveal that Kif15, in contrast to Eg5, does not have a second nucleotide-independent MT-binding site in its C-terminal tail domain. Furthermore, we determine the crystal structure of the binary Kif15-ADP complex captured in the 'ATP-like' state adopted by the switch II cluster and necklinker region.

Determination of protein concentrations
Protein concentrations were determined either by using the Lambert-Beer law measured under native conditions [for the motor-domain constructs bound ADP was taken into account (2500 M À1 cm À1 )] or by applying the Bradford method. The extinction coefficients for each Kif15 construct were calculated (Kif15 1-375 , 17 670 M À1 cm À1 ; Kif15 19-375 , 17 608 M À1 cm À1 ; Kif15 1149-1388 , 10 033 M À1 cm À1 ) using ProtParam from the Bioinformatics Resource Portal (Wilkins et al., 1999).

Analytical gel filtration
Analytical gel filtrations were performed on a Superose 12 10/300 GL column (GE Healthcare). Experiments were conducted at a flow rate of 0.5 ml min À1 (50 mM PIPES pH 6.8, 250 mM NaCl, 2 mM MgCl 2 , 1 mM DTT; all from Sigma) using an injection volume of 200 ml. Prior to running Kif15 expression constructs, the column was calibrated with proteins of known molecular mass (ribonuclease A, 13.7 kDa; carbonic anhydrase, 29 kDa; ovalbumin, 43 kDa; conalbumin, 75 kDa) and blue dextran. The K av values were calculated for the calibration proteins [(V e À V 0 )/(V c À V 0 ), where V e is the elution volume, V 0 is the void volume and V c is the column volume] and were plotted against the log of the molecular masses of the standards. The molecular masses were calculated from the resulting equation.

Steady-state enzyme kinetics
Steady-state basal and MT-stimulated activities were determined using the pyruvate kinase/lactate dehydrogenase coupled assay described previously (Hackney & Jiang, 2001). All measurements were carried out at 25 C using a 96-well Sunrise photometer (Tecan) in a final reaction volume of 100 ml at least in triplicate. Data were analysed using Microsoft Excel 2008 and

Transient-state enzyme kinetics
Slow Mant-ATP [2 0 /3 0 -O-(N-methylanthraniloyl)adenosine-5 0 -triphosphate] turnovers were performed by making manual additions to a 100 ml cuvette in a modified Cary Eclipse fluorescence spectrophotometer at 25 C. Reactions were started by adding the motor to a 5 mM solution of Mant-ATP (Jena Bioscience) in BRB80 (80 mM PIPES-KOH pH 6.9, 1 mM MgCl 2 , 1 mM EGTA) to a final active-site concentration of 3.4 mM. After the plateau of fluorescence had been reached, a 2.5 ml chase of unlabelled 100 mM Mg 2+ -ATP was added. Mant fluorescence was excited at 350 nm and emission was monitored at 450 nm. Stopped-flow experiments were performed at 25 C using a TgK SF-61DX2 stopped-flow spectrofluorimeter. Syringe 1 contained the Kif15 19-375 or Kif15 1-375 single head-Mant-ADP complex, which was formed by adding stock 5 mM Mant-ATP to 2 mM Kif15 in BRB80 buffer with rapid mixing and incubating on ice for 30 min. Syringe 2 contained MTs at various concentrations plus 2 mM Mg 2+ -ATP chasing nucleotide. MTs for these experiments were assembled from pig brain tubulin in BRB80, stabilized using 20 mM taxol and supplemented with 2 mM Mg 2+ -ATP immediately prior to use. Tubulin was dissolved in BRB80 without nucleotide or taxol. Mant fluorescence was excited at 350 nm and emission was monitored at 450 nm. Data were fitted to single exponentials to yield an apparent rate of Mant-ADP release (k off in s À1 ) and an amplitude. Collected data for MT activation and tubulin activation of Mant-ADP release were fitted to rectangular hyperbolas using KaleidaGraph. Data points represent the averages of at least three pushes on at least two separate occasions.

Tubulin purification and polymerization into MTs
Tubulin was purified from bovine brain by cycles of polymerization and depolymerization as described previously (Castoldi & Popov, 2003). Subsequently, tubulin was aliquoted at 18.5 mg ml À1 , snap-frozen and stored in liquid nitrogen. MTs were prepared at 60 mM. The MT concentration was calculated as the heterodimer concentration (110 kDa), assuming that all of the tubulin polymerized. Tubulin was gently mixed with pre-warmed G-PEM buffer (100 mM PIPES pH 6.9, 1 mM EDTA, 1 mM MgCl 2 , 1 mM GTP; all from Sigma) supplemented with 20 mM taxol. Polymerization was carried out overnight at 37 C. For nucleotide binding and ATPase assays MTs were used in the taxol-polymerized form prepared without GTP.

Pelleting assays
Pelleting assays were performed in 50 mM PIPES pH 6.9, 20 mM NaCl, 2 mM MgCl 2 , 1 mM DTT. Kif15 19-375 (1-12 mM) was gently mixed with taxol-polymerized MTs (5 mM) and supplemented with buffer to a final volume of 40 ml. Binding assays were performed in the presence of 2 mM Mg 2+ -ATP, 2 mM AMP-PNP or 4 mU apyrase. Samples were incubated at room temperature for 10 min and centrifuged in a Beckman TL100 ultracentrifuge at 100 000g and 25 C for 20 min. Supernatants were removed from the pellets. The pellet was resuspended in Laemmli reducing sample buffer. Supernatants and pellet samples were run on gradient SDS-PAGE gels (4-12%). All measurements were performed in triplicate. Data were analyzed using ImageJ 143.
B max corresponds to the maximal number of interacting sites.

Protein crystallization, data collection, processing and refinement
Initial nanodrop screening was performed with purified Kif15 19-375 at 10 mg ml À1 in the presence and absence of Mg 2+ -ATP at 4 and 19 C. The protein crystallized at 19 C in the presence of nucleotide and 200 mM MgCl 2 , 25% PEG 3350, 100 mM Tris-HCl pH 8.5. Seeding further optimized the crystals, together with decreasing the protein concentration and the temperature. Improved crystals were immersed in cryoprotectant solution [20%(w/v) meso-erythritol, 240 mM MgCl 2 , 30% PEG 3350, 120 mM Tris-HCl pH 8.5] and flashcooled in liquid nitrogen. Data were collected to a resolution of 2.7 Å at the European Synchrotron Radiation Facility (ESRF). The structure was solved by molecular replacement with MOLREP (Vagin & Teplyakov, 2010) using the CENP-E structure (PDB entry 1t5c; Garcia-Saez et al., 2004) as a search model. The asymmetric unit contained three copies of Kif15 19-375 with bound Mg 2+ -ADP in the catalytic site positioned with REFMAC5 (Murshudov et al., 2011) by rigid-body and restrained refinement. The structure model was improved using Coot (Emsley & Cowtan, 2004) and further refined using PHENIX (Adams et al., 2010) with noncrystallographic symmetry (NCS) restraints.

In vitro assay for neuronal migration
We used a previously established in vitro culture system to study the migration of rat cerebellar granule neurons Hirotsune et al., 1998). Cerebella were isolated from 6-8-day-old rat pups, triturated to give a single-cell suspension and then transfected with either control or Kif15 siRNA and control EGFP plasmid using an Amaxa electroporator and plated on polylysine-coated plastic dishes in serum-containing medium. The next day, the cells were released by treatment with trypsin and were allowed to stand in a solution overnight after removing the trypsin. During this period the cells formed aggregates, which were plated on glass cover slips coated with laminin in serum-free medium to stimulate migration. For time-lapse imaging, the cover slips were placed in an imaging station consisting of a Zeiss environmental chamber and a Zeiss Observer microscope. Serial images were captured every 2 min for a period of 4 h. To quantify cell movement, the total displacement exhibited by the cell body was measured using the AxioVision software and was divided by the time taken for the displacement.

Biophysical characterization of mammalian Kif15 motor and tail domains
Human Kif15 is a protein of 1388 residues ( Fig. 1a) with an N-terminal motor domain (19-375) followed by a long -helical rod-shaped stalk predicted to form an interrupted coiled coil. A bar diagram of Eg5 is presented in Fig. 1(b). Kif15 does not contain the conventional C-terminal globular domain typical of many other kinesins (e.g. Eg5); it possesses a C-terminal leucine zipper, a common dimerization motif (residues 1359-1380). The C-terminal region has also been shown to contain binding sites for the forkhead-associated (FHA) domain of Ki-67 (Sueishi et al., 2000;Vanneste et al., 2009) and for the nuclear protein TPX2 (targeting protein for Xklp2; Tanenbaum et al., 2009) (Fig. 1a). The internal region of Kif15 contains a putative myosin tail homology domain, which has been shown to enable Kif15 to co-localize with actin (Buster et al., 2003).
We subcloned, expressed and purified the Kif15 motor (Kif15 19-375 , coding for residues 19-357) and Kif15 1-375 including the first 18 residues that are thought to include the cover strand, a short nine-residue N-terminal region shown to interact with the C-terminal neck-linker region. These two regions have been shown to fold into the socalled neck-cover bundle by forming a small -sheet and represent a forcegenerating element in several members of the kinesin superfamily Khalil et al., 2008). Finally, we cloned the C-terminal tail construct (named Kif15 1149-1388 ; Fig. 1c  Schematic representation of human Kif15 and Eg5. (a) The Kif15 motor domain shown in green is followed by a discontinuous -helical region predicted to form a coiled-coil stalk domain (coloured blue). The C-terminal leucine-zipper motif is shaded in yellow. The putative interacting regions for actin (residues 743-1333), Ki-67 (residues 1017-1237) and TPX2 (residues 1359-1380) in the C-terminal half of Kif15 are also indicated. (b) The Eg5 motor domain shown in green is followed by a discontinuous -helical region forming a coiled-coil domain (coloured blue). The C-terminal globular tail is marked in pink. (c) Bar diagrams of the Kif15 constructs used in this study. filtration profile for Kif15 1149-1388 was heterogeneous, indicating possible assembly into higher oligomers and/or partial aggregation.

Crystal structure of the binary Kif15 19-375 -Mg 2+ -ADP complex
The structure of the Kif15 motor domain was solved by molecular replacement and refined to a resolution of 2.7 Å . Data-collection and refinement statistics are shown in Table 1. Kif15 19-375 crystallized in space group P3 2 21 with three molecules in the asymmetric unit. The crystals have a Matthews parameter of 2.4 Å 3 Da À1 and a solvent content of 49.1%. The r.m.s. deviations from ideal geometry were 0.013 Å for bond lengths and 1.60 for bond angles for the refined structure. The final model contains residues 24-375 for molecules A and B and residues 24-372 for molecule C. Owing to missing or non-interpretable electron density, several smaller loops could not be built and are absent from the model. Side chains for which no density was observed were deleted from the C position onwards. The Ramachandran plot shows that 94.3% of the residues are in favoured regions, 5.5% are in additionally allowed regions and 0.2% are outliers. Since molecule A presented better defined electron density and a more complete model than molecules B and C, we subsequently use it for further descriptions of the structure.
The Kif15 motor domain shows the typical kinesin fold with an eight-stranded -sheet surrounded by three major -helices on each side (Fig. 2a). The crystallized Kif15 construct contains the neck linker, which aligns with the catalytic core, forming a small -sheet. Although Mg 2+ -ADP is bound in the catalytic site, Kif15 is in the so-called 'ATP-like' conformation. The switch II cluster (helix 4-loop L12-helix 5) is in the so-called 'upward' position, allowing the neck linker to dock to the motor domain (Fig. 2a). We subsequently compared the structure of the Kif15 motor with that of the functionally related Eg5 (Blangy et al., 1995;Turner et al., 2001), using the AMP-PNP-bound complex representing the 'ATP-like' state of Eg5 (Parke et al., 2009). Sequence alignment of Kif15 and Eg5 shows their high similarity inside their motor domains, with 41.8% identical, 15.5% strongly similar and 12.8% weakly similar residues, which is also reflected by their structures (Fig. 2b). Outside their motor domains the primary sequence is at best only weakly conserved, with 20.3% identical, 20.2% strongly similar and 9.8% weakly similar residues. Kif15 has a deletion of two amino acids and a single insertion in the loop L1 region. In the case of Eg5 loop L1 is interrupted by helix 0, while in Kif15 loop L1 is longer. Loop L1 is part of the L1-L3-L5 cluster that together with loop L9 mediates the entrance of the nucleotide (Song et al., 2001). In contrast to Eg5, three residues are missing in the Kif15 loop L2 region. Four residues are missing in loop L5, five in loop L8 (loop L8 is implicated in MT-stimulated ADP release; Fourniol & Moores, 2010) and a single residue is missing in the surface loop L10. There is a fiveresidue insertion in loop L6 of Kif15 which is believed to influence the nucleotide affinity (Kollmar & Glö ckner, 2003). In Eg5, loop L5 is known to be important for inhibitor binding, and differences in the length of this motif in Kif15 (in Kif15 this loop is shorter) explain why Eg5-targeting inhibi- tors do not inhibit Kif15 (Liu et al., 2010). In addition, several other key residues involved in inhibitor binding in Eg5, such as Glu116, Arg119, Trp127 and Ala216, are not present in Kif15. Loop L11 is not visible in any of the three molecules in the asymmetric unit.

Steady-state ATPase activity
We subsequently measured the kinetic parameters for Kif15 19-375 and Kif15 1-375 (Fig. 3 and Table 2). To determine the optimal salt concentration, the rates of basal ATPase activity were measured on varying the NaCl concentration from 0 to 275 mM. The highest basal activity was measured at 75 mM NaCl for Kif15 19-375 and at 50 mM for Kif15 1-375 (Fig. 3a). The differences in activity were rather minor and only decreased slowly at higher salt concentrations. The basal ATPase activity was measured using ATP concentrations from 0 to 2 mM. The k cat for the basal ATPase activity of Kif15 19-375 was 0.054 AE 0.001 s À1 , with a K m,ATP of 40.5 AE 5.4 mM (Fig. 3b, red data points). The k cat for the basal ATPase activity of Kif15 1-375 was 0.028 AE 0.0004 s À1 , about twofold lower, with a K m,ATP of 23.0 AE 1.9 mM (Fig. 3b, blue data points), an almost twofold difference but still similar to the construct without the cover strand. Subsequently, the salt dependence of the MTstimulated ATPase activity of Kif15 19-375 and Kif15 1-375 was determined by measuring the rates at MT concentrations from 0 to 10 mM in the presence of 0, 50, 100 or 150 mM KCl for Kif15 19-375 and of 0 or 50 mM KCl for Kif15 1-375 . With increasing salt concentrations the MTstimulated ATPase activity for both Kif15 constructs decreased significantly (Fig. 3c). The k cat for the MT-stimulated ATPase activity was 2.3 AE 0.1 s À1 with a K 0.5,MT of 1.1 AE 0.1 mM for Kif15 19-375 , and k cat was 2.1 AE 0.1 s À1 with a K 0.5,MT of 3.1 AE 0.3 mM for Kif15 1-375 , indicating that there are no differences in the k cat values between these two Kif15 constructs. Finally, the MT-stimulated ATPase activity in the presence of increasing ATP concentrations (from 0 to 2 mM) was determined (Fig. 3d). The K m,ATP values were 33 AE 3 mM for Kif15 19-375 and 109 AE 20 mM for Kif15 1-375 . With the exception of a threefold difference in the K 0.5,MT and K m,ATP values there was no significant difference between these two constructs, indicating that the presence or absence of the cover strand does not significantly change the ATPase characteristics.

Transient-state ATPase activity
Mant-ATP is a fluorescent analogue of ATP that for some  kinesins gives a fluorescent enhancement on binding at the motor active site. Initial slow transient-state turnover assays of Mant-ATP were performed by making manual additions to a fluorimeter cuvette ( Supplementary Fig. S1a 1 ). As can be seen in our crystal structure, Kif15, like other kinesins, purifies with Mg 2+ -ADP in its active site (Fig. 2a). Addition of an excess of Mg 2+ -Mant-ATP initiates replacement of this active-site Mg 2+ -ADP with Mant-ADP at a rate limited by the basal rate constant of Mg 2+ -ADP release from the Kif15 motor domain. We estimate this rate constant to be 0.017 s À1 . Once a plateau of enhanced fluorescence is reached, addition of a 'chase' of  nonfluorescent Mg 2+ -ATP ( Supplementary Fig. S1a) initiates the opposite reaction, in which active-site Mg 2+ -Mant-ADP is replaced with chasing Mg 2+ -ADP. At 0.005 s À1 , the rate constant for Mant-ADP release is approximately threefold slower than that for Mg 2+ -ADP release, indicating that Mant-ADP binds slightly tighter than Mg 2+ -ADP but is nonetheless a reasonably faithful surrogate and reporter for Mg 2+ -ADP release. Preliminary stopped-flow assays were performed to examine the MT activation of Mant-ADP release (Supplementary Fig. S1b). Kif15 was pre-incubated on ice to load the active site with Mant-ATP. The complex was then rapidly warmed to 25 C, loaded into the stopped flow and rapidly mixed with various concentrations of MTs in the presence of an excess of unlabelled chasing MgATP. The stopped-flow data are preliminary, but nonetheless show that MTs activate MgADP release from Kif15 motor domains $1000-fold. The  (Figs. 4a and 4b).
Comparison of the steady-state ATPase kinetic parameters of Eg5 and Kif15 show that they exhibit optimal in vitro activity at higher salt concentrations (Eg5, 150 mM; Kif15, 50-75 mM). The two motors have similar K m and k cat values for their basal activity, but the MT-activated ATPase activity is higher in the case of Eg5 than of Kif15. The affinities of the motor domains for MTs vary similarly according to the type of bound nucleotide (Lockhart & Cross, 1996), but steady-state, transient-state and direct binding measurements all indicate that the K 0.5,MT is considerably higher in the case of Kif15.

Kif15 contains a nucleotide-dependent MT-binding site in its motor domain, but not in its tail domain
We then determined the affinity of Kif15 19-375 for MTs in the presence of Mg 2+ -ATP, AMP-PNP and apyrase (thus establishing the 'nucleotide-free' state) using the MT pelleting assay. The results for the Kif15 motor domain are shown in Fig. 4 and the K d values as well as the stoichiometry of the Kif15-MT complex are summarized in Table 3. In the presence of the slowly hydrolysable ATP analogue AMP-PNP or apyrase, Kif15 19-375 shows comparable affinities for MTs, with K d values of 0.5 AE 0.2 and 0.4 AE 0.2 mM, respectively, and a stoichiometry close to one (1.1 and 1.3, respectively), indicating that one Kif15 motor domain binds to one -tubulin heterodimer, in agreement with the results reported for other kinesins. Under the conditions tested the Kif15 motor domain shows no affinity for MTs in the presence of Mg 2+ -ATP (in the presence of MTs ATP is rapidly hydrolysed to ADP, which is known to possess low affinity for MTs). In conclusion, human Kif15 contains a nucleotide-dependent MT-binding region in its motor domain.
We then investigated whether or not Kif15 contains a nucleotide-independent MT-binding region in its tail domain (Fig. 4c). A second MT-binding site has been identified in a variety of kinesin tail domains (Liao et al., 1994;Germani et al., 2000;Kuriyama et al., 1994;Echard et al., 1998;Chandra et al., 1993;Karabay & Walker, 1999;Meluh & Rose, 1990;Narasimhulu & Reddy, 1998;Jiang et al., 2007) and has been shown in some cases to be physiologically important for their biological functions. In our hands Kif15 1149-1388 does not bind to MTs in pelleting assays, indicating that the tail does not contain an MT-binding side. This in contrast to human Eg5, which has recently been shown to contain a nucleotide-independent MT-binding region in its tail which is implicated in MT cross-linking and sliding (Weinger et al., 2011).

Functional comparison of Kif15 and Eg5 in a non-mitosis scenario
Given that the mitotic spindle is a rather unique MT apparatus, we wished to investigate whether these two different motors could accomplish similar functions in a nonmitosis scenario. It has previously been reported that Eg5 as well as Kif15 are both enriched in the populations of neurons undergoing migration in the developing brain (Ferhat, Kuriyama et al., 1998;Buster et al., 2003). A role for Eg5 in   regulating neuronal migration has already been documented such that depletion of Eg5 results in neurons that migrate faster (Falnikar et al., 2011). Here, we tested whether Kif15 plays a similar role by depleting it from migrating cerebellar granule neurons using siRNA followed by time-lapse imaging. We found that neurons depleted of Kif15 typically migrated faster but less consistently, such that after a period of vigorous forward movement a cell either remained stationary or underwent a temporary stationary phase before restarting the next phase of forward movement. Quantification showed that the average displacement exhibited by neurons transfected with control siRNA was 24.3 AE 2.5 mm h À1 , whereas that for neurons transfected with Kif15 siRNA was 63.3 AE 6.1 mm h À1 (p < 0.0001; see Fig. 5a). The average time period during which neurons exhibited consistent forward movement was 2.5 AE 0.32 h for the control siRNA group, whereas that for the Kif15 siRNA group was 0.46 AE 0.09 h (p < 0.0001; see Fig. 5b). This phenotype differs from the Eg5-depletion phenotype in terms of the consistency of migration. For tracings of the movements of individual cell bodies, see Fig. 5(c). These observations are reminiscent of previous studies on axonal growth and guidance, in which depleting each of the two motors produced similar but not identical phenotypes (Liu et al., 2010).

Kif15 and Eg5 in mitosis
The role of Kif15 during bipolar spindle formation in early prometaphase reveals that it may -under certain conditionsbe a functional homologue of Eg5. Although their mechanism of action is clearly different (Fig. 6), both proteins are able to work redundantly to fulfill their similar roles. Homotetrameric Eg5 performs its function because of its unique quaternary structure that allows it to cross-link antiparallel MTs and, with its plus-ended directed motility, to slide them apart to form the bipolar spindle. It has recently been suggested that the mechanistics of Eg5 function are even more complex, with additional MT binding sites located in the tail domains (Weinger et al., 2011). In contrast, Kif15 works as a dimer (Wittmann et al., 1998) by crosslinking parallel kinetochore-MTs (Sturgill & Ohi, 2013) and we demonstrated that it does not contain an MT-binding site within its tail domain. The function of the MT linker is most likely carried out by TPX2 through the C-terminal leucine-zipper region of Kif15 (Wittmann et al., 2000). Owing to their similar functions, Kif15 may be able to take over the role of Eg5 following anti-Eg5 chemotherapy by simply up-regulating protein-expression levels, at least in cell culture (Tanenbaum et al., 2009). Such an interchange ability of the two motors, albeit not perfect, is supported by our observations on neurons, which are nonmitotic, suggesting a broadly applicable theme by which two structurally different motors can be made to work in very similar ways. If the scenario of a possible resistance mechanism were true for certain tumours, co-inhibition of both of motors, Eg5 and Kif15, could be a vital therapeutic approach.
This paper reports the crystal structure of human Kif15, which will be a valuable asset for structure-guided design. The comparison of both motors revealed that despite overall functional similarity, Kif15 and Eg5 have important differences, allowing Kif15 to be impervious to Eg5-targeting drugs.

Figure 6
Comparison of current models for Eg5 and Kif15 MT crosslinking and function. Under physiological conditions homotetrameric Eg5 (blue) interacts with antiparallel MTs through two distinct binding sites in its motor and tail domains (two in the motor and two in the tail for each MT it crosslinks), and slides them apart (Weinger et al., 2011). Dimeric Kif15 (green) works predominantly on K-fibres (Sturgill &Ki-67 and the Kif15-TPX2 complex is unknown.